Inertial Sensor Alignment via Machine Learning Orientation Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inertial data captured by sensors in vehicles is often unaligned with the vehicle's coordinate system, limiting its effective use in applications such as automated driving and navigation systems, due to the independent nature of mobile devices and their sensors.
Innovation Solution
The use of Machine Learning (ML) models to estimate the orientation of inertial sensors with respect to the vehicle, allowing for the alignment of inertial data with the vehicle's coordinate system through training on labeled datasets and real-time adjustment of sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If inertial sensors are placed in vehicles in predefined orientations, then the sensors can capture inertial data independently, but the inertial data becomes unaligned with the vehicle's coordinate system
Solution Approach 1:
The patent introduces an intermediary processing system that includes a coordinate transformation module and alignment algorithm. This intermediary transforms the inertial data from sensor coordinates to vehicle coordinates through mathematical rotation and transformation operations, resolving the misalignment issue while preserving sensor independence
Solution Approach 2:
The patent changes the parameter reference frame by applying coordinate transformation matrices that convert inertial measurements from sensor-mounted coordinate systems to vehicle-mounted coordinate systems. This parameter transformation allows data from independently oriented sensors to be accurately represented in the vehicle's reference frame
2Measurement precision
If manual alignment methods are used for inertial sensors, then alignment can be achieved, but the process becomes time-consuming and complex
Solution Approach 1:
The patent replaces manual mechanical alignment procedures with automated computational algorithms. The system uses computer-based coordinate transformation and data processing to automatically align inertial sensor data with the vehicle coordinate system, eliminating time-consuming manual operations while maintaining or improving alignment accuracy
Solution Approach 2:
The alignment system performs self-alignment through automated algorithms that independently process and transform inertial data without requiring manual intervention. The system automatically identifies coordinate transformations and applies corrections, enabling the data processing to serve itself without external assistance
3Adaptability or versatility
If inertial sensors are integrated in mobile devices, then the system becomes more versatile, but the sensors lack inherent alignment with the vehicle's reference frame
Solution Approach 1:
The patent creates a universal alignment framework that can process inertial data from multiple sensor sources and orientations. The coordinate transformation system is designed to handle various sensor configurations and mounting orientations, making the system versatile while ensuring reliable alignment through standardized transformation procedures
Solution Approach 2:
The system implements feedback mechanisms that continuously monitor and adjust the alignment of inertial data. By comparing transformed coordinates with expected vehicle reference frame values, the system can detect and correct misalignment errors, thereby improving data alignment reliability through iterative refinement
Data Source
AI summary
A method of training Machine Learning (ML) models for aligning inertial data captured by inertial sensors located in dynamic vehicles, comprising receiving inertial data recorded in a plurality of trips of vehicles using inertial sensors placed in the vehicles in one or more predefined orientations with respect to the vehicles, segmenting each of the trips to trip segments, creating labeled training samples each comprising a time-series vector comprising inertial data samples recorded during a time window of a respective trip segment and associated with a label reflecting the predefined orientation of the inertial sensors, training one or more ML models using the labeled training samples to estimate orientation of inertial sensors with respect to vehicles, and outputting the trained ML model(s) for estimating orientation of inertial sensors placed in vehicles with respect to the vehicles for aligning their captured inertial data with respect to the vehicles.